Large language models (LLMs) can generate long-form and coherent text, but
they still frequently hallucinate facts, thus limiting their reliability. To
address this issue, inference-time methods that elicit truthful responses have
been proposed by shifting LLM representations towards learned "truthful
directions". However, applying the truthful directions with the same intensity
fails to generalize across different question contexts. We propose LITO, a
Learnable Intervention method for Truthfulness Optimization that automatically
identifies the optimal intervention intensity tailored to a specific context.
LITO explores a sequence of model generations based on increasing levels of
intervention intensities. It selects the most accurate response or refuses to
answer when the predictions are highly uncertain. Experiments on multiple LLMs
and question-answering datasets demonstrate that LITO improves truthfulness
while preserving task accuracy. The adaptive nature of LITO counters issues
with one-size-fits-all intervention-based solutions, maximizing model
truthfulness by reflecting internal knowledge only when the model is confident.

LITO 提出了一种可学习的干预方法，通过识别特定语境下的最佳干预强度，提高真实性，并在高度不确定时选择最准确的回答或拒绝回答。

LITO: 可学习的真实性优化干预

LITO: Learnable Intervention for Truthfulness Optimization

While large language models (LLMs) have taken great strides towards helping
humans with a plethora of tasks like search and summarization, hallucinations
remain a major impediment towards gaining user trust. The fluency and coherence
of model generations even when hallucinating makes it difficult to detect
whether or not a model is hallucinating. In this work, we explore if the
artifacts associated with the model generations can provide hints that the
generation will contain hallucinations. Specifically, we probe LLMs at 1) the
inputs via Integrated Gradients based token attribution, 2) the outputs via the
Softmax probabilities, and 3) the internal state via self-attention and
fully-connected layer activations for signs of hallucinations on open-ended
question answering tasks. Our results show that the distributions of these
artifacts differ between hallucinated and non-hallucinated generations.
Building on this insight, we train binary classifiers that use these artifacts
as input features to classify model generations into hallucinations and
non-hallucinations. These hallucination classifiers achieve up to 0.80 AUROC.
We further show that tokens preceding a hallucination can predict the
subsequent hallucination before it occurs.

我们通过探究模型生成的输入、输出和内部状态中的指标来检测大型语言模型生成中的幻觉，结果表明这些指标在幻觉生成和非幻觉生成之间存在差异，我们进一步通过训练二分类器使用这些指标作为输入特征来将模型生成划分为幻觉和非幻觉，此二分类器的 AUROC 值达到 0.80，并且我们展示了先前的幻觉中的令牌可以预测随后的幻觉。